pay-per-call-mcp — USDC · x402 · AI Agent Billing | LemonCake
Pay-per-call USDC payment proxy for AI agents. Issue scoped Pay Tokens with hard spending caps and auto-journal every charge to freee / Money Forward / QuickBooks.
README
<div align="center">
agent-payment-mcp — LemonCake
Give your AI agent a wallet — with a kill switch.
JWT-based Pay Tokens + USDC micropayments for autonomous AI agents.
No signup. No API keys.npx agent-payment-mcpboots in Demo Mode instantly.
📚 Docs · 🚀 Quickstart · 📧 Contact
Public website, Discord, and Twitter are launching alongside public beta. For now, please reach out via email — we reply within 1 business day.
<br />

<sub>↑ An agent calling a paid API via a Pay Token. Balance drops in real time. Kill Switch stops it in one click.</sub>
</div>
⚡️ 60-second quickstart
# 1. Register at https://lemoncake.xyz, grab a BUYER_JWT
# 2. Issue a Pay Token (scoped, expiring, revocable)
curl -X POST https://lemoncake.xyz/api/tokens \
-H "Authorization: Bearer $BUYER_JWT" \
-H "Content-Type: application/json" \
-d '{"serviceId":"svc_xxx","limitUsdc":"2.00","sandbox":true}'
# → { "jwt": "<pay_token>", ... }
# 3. Hand it to your agent. Done.
# Claude / Cursor:
npx agent-payment-mcp
# Eliza v2:
npm install eliza-plugin-lemoncake
# Anything else:
curl -X POST https://lemoncake.xyz/api/proxy/svc_xxx/your/endpoint \
-H "Authorization: Bearer <pay_token>" \
-H "Idempotency-Key: $(uuidgen)"
Budget exhausted? 402. Token revoked? 422. No runaway agents. No stolen keys.
What is LemonCake?
LLM agents are getting powerful — but they still can't pay for things autonomously.
LemonCake solves this with Pay Tokens: short-lived JWTs that give an agent a scoped spending limit. The agent calls paid APIs through our proxy, gets charged per call in USDC, and stops automatically when the budget runs out.
You Agent Paid API
│ │ │
├─ issue Pay Token ──▶ │ │
│ limit: $2.00 │ │
│ ├─ call /api/proxy ────▶ │
│ │ Authorization: Bearer <pay_token>
│ │ │
│ │ ◀─ response + charge ─┤
│ │ X-Charge-Id: ch_... │
│ │ X-Amount-Usdc: 0.001 │
│ │ │
│ ├─ (budget exhausted) │
│ │ 402 Payment Required │
│ ✗ agent stops cleanly │
🔌 Integrations
MCP サーバー — agent-payment-mcp
Claude Desktop・Cursor に npx 一発で接続できる公式 MCP サーバー。Demo Mode で env vars 空のまま動作(Wikipedia / FX / httpbin が即試せる)。x402-compatible interface。
npx agent-payment-mcp
旧パッケージ名
lemon-cake-mcpは薄いラッパーとして維持中なので、既存のnpx -y lemon-cake-mcpconfigs もそのまま動きます(stderr に deprecation 通知のみ)。
claude_desktop_config.json に追記するだけ:
{
"mcpServers": {
"pay-per-call": {
"command": "npx",
"args": ["-y", "agent-payment-mcp"],
"env": {
"LEMON_CAKE_PAY_TOKEN": "<Pay Token JWT>",
"LEMON_CAKE_BUYER_JWT": "<Buyer JWT>"
}
}
}
}
| ツール | 説明 |
|---|---|
setup |
認証状態の確認と設定手順のガイド(認証不要) |
list_services |
マーケットプレイスの承認済み API 一覧を取得 |
call_service |
Pay Token で課金 API をプロキシ呼び出し |
check_balance |
USDC 残高・KYC ティアを確認 |
Eliza v2 プラグイン — eliza-plugin-lemoncake
@elizaos/core v2 対応の公式プラグイン。character.plugins に追加するだけで Eliza エージェントが自律決済を実行できます。
npm install eliza-plugin-lemoncake
import { lemonCakePlugin } from "eliza-plugin-lemoncake";
const character = {
name: "MyAgent",
plugins: [lemonCakePlugin],
};
# どちらか一方を設定
LEMONCAKE_PAY_TOKEN=eyJhbGci... # クイックスタート(事前発行トークン)
LEMONCAKE_BUYER_JWT=eyJhbGci... # 本番運用(呼び出しごとに都度発行)
自然言語でそのまま動く:
"LemonCake の demo_search_api を 0.50 USDC で呼び出して"
"serviceId: svc_invoice に 0.10 USDC 支払いを実行して"
| アクション | 説明 |
|---|---|
EXECUTE_LEMONCAKE_PAYMENT |
メインアクション。serviceId + limitUsdc を指定して M2M 決済を実行 |
PAY_WITH_LEMONCAKE / M2M_PAYMENT など |
自然言語トリガー用エイリアス(similes) |
→ 詳細: eliza-plugin-lemoncake/README.md
✨ Features
Packages (5)
| Package | What it does |
|---|---|
agent-payment-mcp |
Pay-per-call USDC for any HTTP API. Demo Mode: Wikipedia / FX / httpbin, no signup. |
xstocks-mcp |
Buy tokenized stocks (AAPLx, TSLAx, SPYx…) via Jupiter DEX on Solana. No broker. |
alpaca-guard-mcp |
Trade US stocks via Alpaca Markets with per-session spend caps and kill switch. |
tokenized-stock-mcp |
Buy Dinari dShares (tokenized US stocks) with USDC. Sandbox default. |
@lemon-cake/mcp-sdk |
SDK for MCP developers — add pay-per-call billing to your own MCP server in 3 lines. |
For AI Agents (Buyers)
- Pay Token (JWT) — Scoped, expiring spend authorization. One token per task or session.
- 402-first design — Agents receive structured
402 Payment Requirederrors with machine-readable codes when budget runs out. - Idempotency keys — Prevent double charges on retries (auto-assigned by plugins).
- Real-time balance — Check remaining USDC before committing to expensive calls.
- Kill Switch — Atomic one-click token revocation. Race-condition-free.
For API Providers (Sellers)
- Service registry — Register any REST API. Set price-per-call in USDC.
- Instant revenue — Get paid per call with no invoicing, no net-30, no chargebacks.
- Usage analytics — See call counts, revenue, and error rates per service.
Infrastructure
- JPYC on-chain deposit — Charge balance with JPYC (Polygon ERC-20). Auto-verified via TX hash.
- KYA (Know Your Agent) — Tiered spending limits: 10 → 1,000 → 50,000 USDC/day.
🏗 Architecture
┌──────────────────────────────────────────────────────────┐
│ lemoncake.xyz │
│ │
│ ┌─────────────┐ ┌──────────────┐ ┌─────────────┐ │
│ │ Dashboard │ │ Hono API │ │ Pay Proxy │ │
│ │ (Next.js) │◀──▶│ + OpenAPI │◀──│ Middleware │ │
│ └─────────────┘ └──────┬───────┘ └─────────────┘ │
│ │ │
│ ┌────────────┼────────────┐ │
│ ▼ ▼ ▼ │
│ PostgreSQL Redis Polygon │
│ (Prisma) (Queue) USDC / JPYC │
└──────────────────────────────────────────────────────────┘
▲ ▲
│ │
MCP / Eliza Upstream APIs
(Claude, Cursor, Eliza...) (registered services)
Key design decisions:
- HMAC-SHA256 (HS256, RFC 7518) for JWT signing today; Ed25519 (EdDSA, RFC 8037) migration planned for v0.1.0 to enable asymmetric verification by external partners
- Optimistic locking on
usedUsdc— prevents double charges under concurrent agent calls - Proxy-first — upstream API keys never leave the server; agents only hold Pay Tokens
🔌 Public API (selected endpoints)
Full OpenAPI spec available at
/api/docafter signing in.
Auth
POST /api/auth/register
{ "name": "string", "email": "string", "password": "string (min 8)" }
→ { "token": "<buyer_jwt>", "expiresIn": 2592000 }
POST /api/auth/buyer-login
{ "email": "string", "password": "string" }
→ { "token": "<buyer_jwt>" }
Tokens (Pay Token)
POST /api/tokens
Authorization: Bearer <buyer_jwt>
{
"serviceId": "<id>",
"limitUsdc": "5.00",
"buyerTag": "my-agent-session-42",
"expiresAt": "2026-05-01T00:00:00Z"
}
→ { "tokenId": "...", "jwt": "<pay_token>", "limitUsdc": "5.000000", "expiresAt": "..." }
Proxy (Pay-per-call)
ANY /api/proxy/<serviceId>/<upstream-path>
Authorization: Bearer <pay_token>
Idempotency-Key: <uuid>
→ upstream response
+ X-Charge-Id: ch_...
+ X-Amount-Usdc: 0.001000
Error responses:
{ "error": "Token limit exceeded", "used": "4.999", "limit": "5.000" } // 402
{ "error": "Insufficient balance: 1.23 USDC available" } // 402
{ "error": "Token expired" } // 401
Services
GET /api/services?reviewStatus=APPROVED&limit=50
→ [{ "id": "...", "name": "...", "pricePerCallUsdc": "0.001", ... }]
🧑💻 Local Development
# 1. Clone
git clone https://github.com/evidai/agent-payment-mcp.git
cd agent-payment-mcp
# 2. API server
cd api
cp .env.example .env
npm install && npx prisma migrate dev
npm run dev # http://localhost:3000
# 3. Dashboard
cd ../dashboard
cp .env.example .env.local # NEXT_PUBLIC_API_URL=http://localhost:3000
npm install && npm run dev # http://localhost:3001
# 4. MCP server
cd ../mcp-server
npm install && npm run build
# 5. Eliza plugin
cd ../eliza-plugin-lemoncake
npm install && npm run build
Seed demo data
node api/seed_demo.js
# Creates demo provider, 4 approved services, and 9,000+ sample charge records
🛡 Security
- Brute-force protection — 10 failed logins triggers a 15-minute lockout per email
- Pay Token scoping — Each token is bound to a single
serviceId; cross-service reuse is rejected - No JWT storage — Only the
jti(token ID) is stored in the DB; the signed JWT never persists - CSRF protection — OAuth state parameter with 10-minute TTL and single-use consumption
- On-chain deposit verification — JPYC transfers are verified against Polygon event logs before balance is credited
🗺 Roadmap
Shipped:
- [x] agent-payment-mcp — USDC pay-per-call for any HTTP API, Demo Mode included
- [x] xstocks-mcp — Jupiter DEX xStocks (AAPLx/TSLAx/SPYx) on Solana mainnet
- [x] alpaca-guard-mcp — Alpaca Markets proxy with spend caps and kill switch
- [x] tokenized-stock-mcp — Dinari dShares with USDC, sandbox default
- [x] @lemon-cake/mcp-sdk — SDK for MCP developers to add Pay Token billing in 3 lines
- [x] Kill Switch — atomic one-click token revocation (race-condition-free)
- [x] KYA (Know Your Agent) — tiered spending limits: 10 → 1,000 → 50,000 USDC/day
- [x] Sandbox / Demo Mode — full-flow dry-run without moving real USDC
- [x] JPYC on-chain deposit with Polygon event-log verification
Next:
- [ ] Panic button — revoke all active tokens for a buyer in one click
- [ ] LemonCake marketplace — register and monetize any REST API
- [ ] Streaming / token-count billing for LLM APIs
- [ ] Agent-to-agent sub-token delegation (parent caps child)
- [ ] Self-hostable edition
📄 License
Proprietary — All rights reserved © 2026 LemonCake
MCP server and Eliza plugin source are available for review. Core API and payment engine are closed source.
Built for the agentic web. lemoncake.xyz
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.